190071B

Bandara D.R.K.W.M.S.D.

Question 1

import numpy as np
import cv2 as cv
import matplotlib.pyplot as plt

img = cv.imread('emma_gray.jpg', cv.IMREAD_GRAYSCALE)
assert img is not None

transform1 = np.linspace(0,50,50)
#transform2 = np.linspace(50,100,1)
transform3 = np.linspace(100,255,100)
#transform4 = np.linspace(255,150,1)
transform5 = np.linspace(150,255,106)
transform = np.concatenate((transform1,transform3,transform5), axis = 0).astype(np.uint8)

transformedImg = cv.LUT(img,transform)
# transformedImg = transform[img]#both are same this is the numpy way of doing that


assert len(transform)==256
im = cv.cvtColor(img, cv.COLOR_BGR2RGB)
im2 = cv.cvtColor(transformedImg, cv.COLOR_BGR2RGB)
fig, ax = plt.subplots(1,3, figsize = (20,20))
ax[0].imshow(im)
ax[1].imshow(im2)
ax[0].set_title('Original')
ax[1].set_title('After transformation')
ax[2].plot(transform)
ax[2].set_aspect('equal')
plt.show()

Question 2

White Matters

import numpy as np
import cv2 as cv
import matplotlib.pyplot as plt

img = cv.imread('brain_proton_density_slice.png', cv.IMREAD_GRAYSCALE)
imgConverted = cv.cvtColor(img, cv.COLOR_BGR2RGB)
fig, ax = plt.subplots(1,3,figsize = (20,20))
ax[0].imshow(imgConverted)


#white matter transformation
transform1 = np.linspace(0,25,210)
#transform2 = np.linspace(50,100,1)
transform3 = np.linspace(25,240,20)
#transform4 = np.linspace(255,150,1)
transform5 = np.linspace(240,255,26)
transform = np.concatenate((transform1,transform3,transform5), axis = 0).astype(np.uint8)
transformedImg = cv.LUT(img,transform)
imgConvertedTransformed = cv.cvtColor(transformedImg, cv.COLOR_BGR2RGB)
ax[2].imshow(imgConvertedTransformed)
ax[1].plot(transform)
ax[1].set_aspect('equal')
ax[0].set_title('Original Image')
ax[2].set_title('White matters')
ax[1].set_title('Transformation for white matters')
plt.show()

Gray Matters

#gray matter transformation
transform1 = np.linspace(0,5,180)
transform2 = np.linspace(5,225,5)
transform3 = np.linspace(225,225,26)
transform4 = np.linspace(225,5,5)
transform5 = np.linspace(5,5,40)


fig, ax = plt.subplots(1,2, figsize = (10,10))

transform = np.concatenate((transform1,transform2,transform3,transform4,transform5), axis = 0).astype(np.uint8)
transformedImg = cv.LUT(img,transform)
imgConvertedTransformed = cv.cvtColor(transformedImg, cv.COLOR_BGR2RGB)
ax[0].plot(transform)
ax[0].set_aspect('equal')
ax[1].imshow(imgConvertedTransformed)
ax[1].set_title('Gray matters')
ax[0].set_title('Transformation for Gray matters')
plt.show()

Discussion

White matters and gray matters can be accentuate using linear transformations. Using linear tarnsformations, map the pixel values of the image such that different features and colors are accentuate.

Question 3

import numpy as np
import cv2 as cv
import matplotlib.pyplot as plt

img = cv.imread('highlights_and_shadows.jpg', cv.IMREAD_COLOR)
imgHab = cv.cvtColor(img, cv.COLOR_BGR2LAB)
gamma = [0.5]
for k in gamma:
    transform = np.array([(p/255)**k*255 for p in range(0,256)]).astype(np.uint8)#
    imgHab[:,:,0]  = cv.LUT(imgHab[:,:,0],transform)
    convertedImg1 = cv.cvtColor(img, cv.COLOR_BGR2RGB)
    convertedImg2 = cv.cvtColor(imgHab, cv.COLOR_LAB2RGB)
    fig, ax = plt.subplots(5,1, figsize = (20,20))
    ax[0].plot(transform)
    ax[0].set_title('Gamma = %s'%k)
    ax[0].set_aspect('equal')
    ax[1].imshow(convertedImg1)
    ax[2].imshow(convertedImg2)
    ax[1].set_title('Original')
    ax[2].set_title('Gamma = %s'%k)
    hist_img1 = cv.calcHist([img],[0],None,[256],[0,256])
    hist_img2 = cv.calcHist([imgHab],[0],None,[256],[0,256])
    ax[3].plot(hist_img1)
    ax[4].plot(hist_img2)

plt.show()

Discussion

intensity of the image has increased after gamma correction. Also the histogram of the equalized image is distributed over the full range. But in original image, histogram is compressed to lower pixel values. By distributing this, we can increase the brightness of the image.

Question 4

import numpy as np
import cv2 as cv
import matplotlib.pyplot as plt

img = cv.imread('shells.png', cv.IMREAD_GRAYSCALE)
print(img.shape)
imgConverted = cv.cvtColor(img, cv.COLOR_BGR2RGB)
fig, ax = plt.subplots(2,1, figsize = (10,10))
hist , bins = np.histogram(img.ravel() , 256 , [0,256])
ax[0].plot(hist)
EqualizeMap = []
sum1 = 0
for i in range(len(bins)-1):
    sum1 += hist[i]
    a = int(round(sum1*255/(img.shape[0]*img.shape[1]),1))
    EqualizeMap.append(a)
    
EqualizedImage = cv.LUT(img,np.array(EqualizeMap).astype(np.uint8))
imgConverted2 = cv.cvtColor(EqualizedImage, cv.COLOR_BGR2RGB)
hist_equalized , bins_equalized = np.histogram(EqualizedImage.ravel() , 256 , [0,256])
ax[1].plot(hist_equalized)
fig, ax = plt.subplots(1,2, figsize = (10,10))
ax[0].imshow(imgConverted)
ax[1].imshow(imgConverted2)

(500, 500)

<matplotlib.image.AxesImage at 0x220561d0>

Discussion

intensity of the image has increased after equalization. Also the histogram of the equalized image is distributed over the full range. But in original image, histogram is compressed to lower pixel values. By distributing this, we can increase the brightness of the image.

Question 5

import numpy as np
import cv2 as cv
import matplotlib.pyplot as plt
########################################
#custom zooming functions
########################################
def zoom(img,factor):
    zoomimg = np.zeros((int(img.shape[0]*factor),int(img.shape[1]*factor),3),dtype = img.dtype)
    for i in range(zoomimg.shape[0]):
        for j in range(zoomimg.shape[1]):
            x = int(round(i/factor,0))
            y = int(round(j/factor,0))
            if x >= img.shape[0]:
                x = img.shape[0]-1
            if y >= img.shape[1]:
                y = img.shape[1]-1
            zoomimg[i,j] = img[x,y]
    return zoomimg

def zoomBilinear(img,factor):
    zoomimg = np.zeros((int(img.shape[0]*factor),int(img.shape[1]*factor),3),dtype = img.dtype)
    for i in range(zoomimg.shape[0]):
        for j in range(zoomimg.shape[1]):
            x = i / factor
            y = j / factor
            x_1 = int(x)
            y_1 = int(y)
            x_2 = int(x) + 1
            y_2 = int(y) + 1
            if x_1 >= img.shape[0]:
                x_1 = img.shape[0]-1
            if x_2 >= img.shape[0]:
                x_2 = img.shape[0]-1
            if y_1 >= img.shape[1]:
                y_1 = img.shape[1]-1
            if y_2 >= img.shape[1]:
                y_2 = img.shape[1]-1
            value_1 = img[x_1,y_1]*abs(y-y_2) + img[x_1,y_2] * abs(y-y_1)
            value_2 = img[x_2,y_1]*abs(y-y_2) + img[x_2,y_2] * abs(y-y_1)
            value_3 = value_1*abs(x-x_2) + value_2*abs(x-x_1)
            for k in range(3):
                value_3[k] = int(round(value_3[k],0))
            zoomimg[i,j] = value_3
    return zoomimg


img = cv.imread('a1q5images\im02small.png')
imgConverted = cv.cvtColor(img, cv.COLOR_BGR2RGB)
fig, ax = plt.subplots(1,3, figsize = (30,30))
ax[0].imshow(imgConverted)
ax[0].set_title('Original')
zoomImgbilinear = zoomBilinear(img, 4)
zoomImgNearestNeighbour = zoom(img, 4)
        
imgConverted2 = cv.cvtColor(zoomImgbilinear, cv.COLOR_BGR2RGB)
ax[1].imshow(imgConverted2)
ax[1].set_title('Bilinear Zoom')
cv.imwrite("zoomedBiliniear.jpg",zoomImgbilinear)

imgConverted3 = cv.cvtColor(zoomImgNearestNeighbour, cv.COLOR_BGR2RGB)
ax[2].imshow(imgConverted2)
ax[2].set_title('Nearest neighbour Zoom')
cv.imwrite("zoomed.jpg",zoomImgNearestNeighbour)

img1 = cv.imread('a1q5images\im02.png')
print("SSD of Nearest neighbour method",sum(sum((img1 - zoomImgNearestNeighbour)**2))/img1.shape[0]*img1.shape[1])
print("SSD of bilinear method", sum(sum((img1 - zoomImgbilinear)**2))/img1.shape[0]*img1.shape[1])

SSD of Nearest neighbour method [331.2 396.8 179.2]
SSD of bilinear method [363.2   8.  145.6]

Discussion

The zoomed image using bilinear interpolation has good quality more then zoomed using nearest neighbourhood method because that method is more realistic than nearest neighbourhood method.

Question 6

import numpy as np
import cv2 as cv
import matplotlib.pyplot as plt

################
#Custom Sobel function
def sobel(img, kernal):
    kernal_w = kernal.shape[0]
    kernal_h = kernal.shape[1]
    padsize1 = kernal_w//2
    padsize2 = kernal_h//2

    centerPointx = kernal_w//2 + 1
    centerPointy = kernal_h//2 + 1

    filteredImg = np.zeros((img.shape[0],img.shape[1]), dtype = np.float32)
    for i in range(padsize1,img.shape[0]-padsize1):
        for j in range(padsize2, img.shape[1]-padsize2):
            mat1 = img[i-kernal_w//2:i + kernal_w//2+1, j-kernal_h//2:j + kernal_h//2+1]
            value = sum(sum(mat1 * kernal))
            filteredImg[i,j] = value
    return filteredImg
########################################################33

img = cv.imread('einstein.png',cv.IMREAD_GRAYSCALE)
imgConverted = cv.cvtColor(img, cv.COLOR_BGR2RGB)
fig, ax = plt.subplots(1,6, figsize = (30,30))
ax[0].imshow(imgConverted)
ax[0].set_title('Original Image')
############################################
###inbuilt sobel
Sobel_V = np.array([[-1,-2,-1],[0,0,0],[1,2,1]],dtype = np.float32)
Sobel_H = np.array([[1,0,-1],[2,0,-2],[1,0,-1]],dtype = np.float32)
imgWithSobelV = cv.filter2D(img,-1,Sobel_V)
imgWithSobelH = cv.filter2D(img,-1,Sobel_H)
###########################################
#Call the custom sobel function
###################################
imgCustomSobelV = sobel(img, Sobel_V)
imgCustomSobelH = sobel(img, Sobel_H)
#####################################
ax[1].imshow(imgWithSobelV, cmap = 'gray',vmin = 0,vmax = 255)
ax[1].set_title('In built vertical sobel')
ax[2].imshow(imgWithSobelH, cmap = 'gray',vmin = 0,vmax = 255)
ax[2].set_title('In built horizontal sobel')
ax[3].imshow(imgCustomSobelV, cmap = 'gray',vmin = 0,vmax = 255)
ax[3].set_title('Custom  vertical sobel')
ax[4].imshow(imgCustomSobelH, cmap = 'gray',vmin = 0,vmax = 255)
ax[4].set_title('Custom horizontal sobel')
sobel_1 = np.array([[1],[2],[1]],dtype = np.float32)
sobel_2 = np.array([[1, 0, -1]],dtype = np.float32)
img_1 = sobel(img, sobel_1)
img_2 = sobel(img_1, sobel_2)
ax[5].imshow(img_2, cmap = 'gray',vmin = 0,vmax = 255)
ax[5].set_title('Sobel filter using Multiplication property')

Text(0.5, 1.0, 'Sobel filter using Multiplication property')

Discussion

In built filter 2D functions and custom sobel filter functions give same results. Also 1D sobel filtering using mutiplication property is also give same results as prevoius

Question 7

import numpy as np
import cv2 as cv
import matplotlib.pyplot as plt

img = cv.imread('daisy.jpg')
imgConverted = cv.cvtColor(img, cv.COLOR_BGR2RGB)
fig, ax = plt.subplots(1,4, figsize = (20,20))
ax[0].imshow(imgConverted)
ax[0].set_title('Original')
mask = np.zeros(img.shape[:2],np.uint8)
background = np.zeros((1,65),np.float64)
foreground = np.zeros((1,65),np.float64)
rect = (50,120,520,700)
cv.grabCut(img,mask,rect,background,foreground,5,cv.GC_INIT_WITH_RECT)
mask2 = np.where((mask==2)|(mask==0),0,1).astype('uint8')
img_1 = img*mask2[:,:,np.newaxis]
imgConverted2 = cv.cvtColor(img_1, cv.COLOR_BGR2RGB)
ax[1].imshow(imgConverted2)
ax[1].set_title('Foreground')
img_2 = img - img_1
imgConverted3 = cv.cvtColor(img_2, cv.COLOR_BGR2RGB)
ax[2].imshow(imgConverted3)
ax[2].set_title('Background')
imgConverted4 = cv.cvtColor(mask, cv.COLOR_BGR2RGB)
ax[3].imshow(mask2, cmap = 'gray')
ax[3].set_title('Mask')

Text(0.5, 1.0, 'Mask')

Image Enhancing

import numpy as np
import cv2 as cv
import matplotlib.pyplot as plt

img = cv.imread('daisy.jpg')
imgConverted = cv.cvtColor(img, cv.COLOR_BGR2RGB)
fig, ax = plt.subplots(1,2, figsize = (20,20))
ax[0].imshow(imgConverted)
ax[0].set_title('Original')
img_3 = cv.GaussianBlur(img_2, (9,9),0)
img_4 = img_1 + img_3
imgConverted5 = cv.cvtColor(img_4, cv.COLOR_BGR2RGB)
ax[1].imshow(imgConverted5)
ax[1].set_title('Enhanced Image')

Text(0.5, 1.0, 'Enhanced Image')

Discussion

Background of just beyond the edge of the flower quite dark becuase running kernal of the gaussian blur function for the background spreads the black color around the dark edges of the background image. This problem can be solve using gaussian filter to original image and replace foreground with non blured one

GitHub Link https://github.com/Saliya-99/In19-S4-EN2550---Fundamentals-of-Image-Processing-and-Machine-Vision/commits/master/190071B.ipynb